Accelerator

Instinct MI355X

AMD CDNA 4 oam accelerator summary for training, inference, and roofline-style performance analysis.

Vendor

AMD

Architecture

CDNA 4

Unit

OAM accelerator

Form factor

OAM module

Launch

2025-06-12

Memory

288 GB HBM3E

HBM bandwidth

8 TB/s

BF16 peak

2.5 PFLOPS

FP16 peak

2.5 PFLOPS

FP8 dense peak

5 PFLOPS

FP8 sparse peak

10.1 PFLOPS

FP4 dense peak

10.1 PFLOPS

FP4 sparse peak

n/a

FP64 peak

78.6 TFLOPS

INT8 peak

5 POPS

Interconnect

AMD Infinity Fabric - 7 links, 153 GB/s peak per link

Power

1400 W TBP

Software stack

ROCm

Notes

CDNA 4 part with native MXFP4 and MXFP6 support for low-precision AI workloads.
The 8-GPU MI355X platform is listed with 2.3 TB total HBM3E and 64 TB/s aggregate HBM bandwidth.